Machine Fault Diagnosis And Performance Degradation Assessment Using Wavelet Packet Entropy And Gaussian Mixture Model

Periodic maintenance is on-condition, which is hard to meet the needs of recentproduction. Therefore, the active development of maintenance strategy according to therunning status is being on the agenda.The performance degradation assessment of equipment is a maintenance technologybased on the active maintenance strategy.This technology is different from the traditionalfault diagnosis methods because of its focus on the prediction and analysis of the equipmentlife within its whole life cycle instead of in a particular state. This thesis takes the rollingbearing as the research object and extracts the wavelet package entropy feature of its vibrationsignals. Based on the feature, the benchmark state and operation state of Gaussian MixtureModel (GMM) are also established. Besides, the equipment performance degradation state isanalyzed through comparing the deviation of operation state relative to the benchmark state.The thesis mainly includes three parts.According to the characteristics of the feature extraction based on the waveletdecomposition and the shock feature of the fault signal, the appropriate wavelet packet basisand decomposition level are selected.GMM is applied to diagnosis rolling bearing and compared with the SVM. The resultindicates that the proposed method was effective and accurate.The concept of Deviate Value (DV) is presented and used to analyze the performancedegradation of rolling bearing based on the simulation data. The result is compared with thetime domain feature. In this method, the degradation data of rolling bearing within the wholelife cycle is also analyzed and compared with the equipment performance degradation methodbased on logistic regression. The result indicates that the proposed method has bettercapability in early fault identification and accurately reflects the whole performancedegradation process of the rolling bearing without historical data and the definition of prioriprobability.